In 2016 DeepMind built a system that recommended cooling adjustments for Google’s data centres, but human operators still had to review and apply each suggestion. On August 17, 2018, DeepMind announced the next step: the AI now controls the cooling directly, choosing and implementing actions on its own while staying under the supervision of data-centre staff. This shift from an advisory tool to an autonomous controller is what made the milestone notable.
Every five minutes, a cloud-based system pulls a snapshot of the cooling plant from thousands of sensors, feeds it into deep neural networks that predict how different combinations of actions will affect future energy use, and selects the actions that minimise energy while respecting safety constraints. A local control system at the data centre verifies each instruction against its own limits before acting, and human operators can override the AI or exit autonomous mode at any time. According to DeepMind, the system delivered consistent energy savings of around 30 percent on average, improving from a 12 percent saving at launch to roughly 30 percent over nine months of operation as it accumulated data.
For a business reader, this is an early, concrete example of AI being trusted to run a safety-critical industrial process rather than just advise on it, with clear guardrails - a pattern that recurs across manufacturing and energy operations.